Daily Brief

Full episode: https://www.youtube.com/watch?v=f0J_EeW42Lo Me on twitter: https://x.com/dwarkesh_sp

Code is being written faster than ever, and Nick Nisi, who leads developer and agent experience at WorkOS, argues that the human reviewer is now the bottleneck. To make the point, he hooked a simulated fruit fly brain, built from a full scan of its neurons, up to code review: AI turns each change into literal "code smells," sugar and bitter rewards teach the fly what good and bad code taste like, and it was right about 88% of the time in testing. Then he plays a live round against the audience and lays out a practical plan where team review skills, classifiers and automated reviewers only send

AI has moved the bottleneck from writing code to reviewing it, and SREs are left handling the incidents. Andrew Norrish of Sourcegraph explains why unreviewed AI code is driving more outages, and why SRE and platform teams are best placed to fix it. He also shows how code search and a code-graph context engine help teams investigate, prevent and fix issues across huge codebases. 0:00 Intro 1:28 The story of AI-driven development 2:52 Code review is the new bottleneck 4:27 Why hiring more SREs isn't enough 7:06 What the industry data shows 9:20 Why SREs and platform teams matter more now 13:14

A coding agent tried to fix a staging problem and deleted a production database and its backup instead. It did this with valid credentials. Sagar Batchu of Speakeasy uses that incident to lay out a control layer for agents in regulated industries. Every action is tied to an identity, checked against policy in real time, never allowed to exceed its owner's rights, and recorded for compliance. 0:00 Intro 1:21 The agent that deleted production 3:43 How this is playing out across the industry 4:48 What we already expect from people 5:26 A control layer for agents 8:26 Connect, inspect, decide, re

Agents relearn the same things every session because the knowledge that runs a company never gets written down. Will Brown of Prime Intellect makes the case for training models on your own traces, feedback and workflows, so they learn on the job the way human coworkers do. He covers evals as test cases, decision models and routing, and widget-style interfaces that turn your decisions into training data. He ends with nightly training runs and an example model trained for about $300K. 0:00 Intro 1:21 From chatbots to AI-native organizations 2:33 What Prime Intellect builds 3:45 The knowledge th

Runway's first AI film in 2019 was 128x128 pixels. Today, people can't tell generated video from real footage. Runway co-founder and CEO Cristóbal Valenzuela explains how next-frame prediction at scale teaches models physics, object permanence and how the world works. He then shows what happens when those models become real-time and interactive: world interface models, robotics, and Continuum, an early research prototype where anything on your screen can become a live, interactive video. 0:00 Intro 0:50 Runway's first AI film (2019) 1:49 State-of-the-art video today 2:48 Video as next-frame p